I build practical software at the point where marketing, automation and AI meet. I like projects where a business problem comes first and the code is there to solve it: a lead that never gets lost, a weekly report that writes itself, a brand audit that takes minutes instead of days.
Below are four projects, grouped by what you might be looking for.
claude-brief-cli - A TypeScript command-line tool that turns messy notes into a structured brief (summary, decisions, action items with owners) using the Claude API. Strict types, zod-validated model output, explicit error handling, tests written first, and CI.
hermes-agent-ops - A write-up of running a self-hosted AI agent stack on a small VPS: orchestration, scheduled jobs, a security hardening pass, and an honest postmortem of what went wrong and how it was fixed.
brand-equity-dashboard - A small web app that scores a brand on Keller's CBBE pyramid, draws radar and pyramid charts, and turns weak layers into plain-English recommendations. It exports the audit as markdown. Responsive and keyboard-friendly, with the scoring logic fully unit-tested.
n8n-automation-pack - Three importable n8n workflows: lead capture to a spreadsheet and Slack alert, a weekly lead digest, and signup enrichment with hot-lead alerts. Webhooks are HMAC-verified, and every workflow has an error branch.
- Start from the problem, then pick the tools.
- Make failures visible: explicit errors, alerts, and tests for the unhappy paths.
- Keep secrets out of code and say plainly what has and hasn't been tested.
TypeScript, Node.js, Vite, Vitest, ESLint, GitHub Actions, n8n, the Claude API, Linux and systemd basics, Tailscale.
Open an issue on any repo above, or message me through my GitHub profile.